What Is StarryLink Optical Module?

In the AI era, data center network interconnection presents new challenges for optical modules, requiring significant improvements in transmission distance, O&M efficiency, and interconnection security. To address these demands, Huawei has launched the StarryLink optical module brand. And to keep up with the rapid growth of AI computing power, Huawei offers StarryLink optical modules that can be sold separately, compatible with various types of computing NICs and switches. Through rigorous quality control and end-to-end R&D and manufacturing, Huawei's StarryLink optical modules deliver high-quality interconnection that supports long-term business development.

Contents

What Is StarryLink Optical Module?

At MWC 2025, Huawei officially launched the StarryLink optical module to the global market. The StarryLink optical module series is designed to deliver a premium "3S" network experience—Spanning (ultra-long-distance transmission), Stable (exceptional reliability), and Secure (enhanced security)—to accelerate enterprise digital and intelligent transformation.

Why Do We Need StarryLink Optical Module?

With the rapid advancement of AI, the scale of AI training clusters is expanding from thousands to tens of thousands of GPUs. In this transformation, network interconnection plays a critical role in unleashing the full potential of system-wide computing power. Meanwhile, the exponential increase in optical module deployment introduces unprecedented challenges in reliability and maintenance.

The following uses large-scale model training as an example:

Training a GPT-3-scale cluster, typically comprising around 1000 GPUs, requires approximately 2500 x 200G optical modules or 4000 x 400G optical modules.
For a GPT-4-scale cluster involving around 10,000 GPUs, approximately 25000 x 200G optical modules or 40000 x 400G optical modules are required to support high-speed interconnection.

On average, each additional GPU requires 2.5 to 4 optical modules to support network communication. Deploying such a vast number of modules makes network system stability a critical factor in maximizing computing performance.

However, in real-world operations, the failure rate of optical modules is not negligible. Based on an industry-average annual failure rate of 4‰, a 10,000-GPU cluster may experience one interruption every 3.6 days due to optical module failures, with each recovery taking approximately 2 hours. The resulting compute downtime alone could lead to an economic loss of up to CNY1.4 million per day. High-reliability, high-performance StarryLink optical modules are ideally suited to address the challenges of network reliability and maintenance in the era of 10,000-GPU clusters, and to improve training efficiency while minimizing failure-related losses.

Key Technologies of StarryLink Optical Modules

Designed for data center and AI network interconnection scenarios, Huawei's StarryLink optical modules offer three core advantages:

Ultra-long-distance transmission: Huawei's optical path coupling algorithm uses deep optical design to control and optimize the power distribution of optical signals, ensuring maximum signal strength and extending the transmission distance. This helps avoid costly equipment room cabling upgrades and significantly reduces interconnection costs.
Exceptional reliability: Huawei's short-distance optical return loss positioning technology detects contamination or connector looseness by analyzing reflected signals caused by obstructions in the optical links. By comparing the intensity and latency of return signals against normal signals, the technology accurately identifies fault points, improving overall module reliability by an order of magnitude and considerably lowering failure rates.
Enhanced security: Huawei's PHYSec physical-layer encryption technology enables 100% packet encryption at the optical digital signal processor (oDSP) level. This offloads upper-layer processing and ensures stronger user data privacy protection. The technology delivers low latency, zero bandwidth loss, and flexible deployment, providing full-link, hardware-level encryption for data centers and enabling a higher level of network security assurance.

Application Scenarios of StarryLink Optical Modules

Huawei's StarryLink optical modules are designed for next-generation data centers and intelligent computing networks. They are ideally suited for high-speed interconnection scenarios such as AI training clusters, high-performance computing (HPC), financial cloud platforms, and large-scale enterprise data centers. The modules focus on enhancing network reliability, transmission distance, and data security to deliver the high-bandwidth and low-latency interconnection demanded by large-scale compute clusters.

For example, in NVD scenarios, StarryLink optical modules can be deployed for network interconnection in the spine-leaf architecture. Taking the XH9230 51.2T switch as an example, multiple 400G QSFP112 module types, including FR4, DR4, SR4, and VR4, can be used between spine and leaf nodes to accommodate various link distances and transmission modes. Between leaf nodes and servers, multi-rate combinations such as 400G OSFP-RHS and 200G QSFP112 are supported, matching mainstream NICs like CX7 and BF3 to enable high-speed interconnection across diverse server types. This solution supports large-scale GPU cluster deployments, ensuring efficient compute distribution and robust communication performance within the cluster.

In a real-world deployment for a financial enterprise, Huawei's StarryLink optical modules delivered outstanding results. As part of the company's digital transformation, its data center network evolved from 40G to 100G, with over 1000 racks deployed across multiple equipment rooms, placing higher demands on interconnection distance and signal quality. Huawei provided a 100G QSFP-BIDI multimode optical module solution, enabling reliable 150-meter transmission with no changes required to the existing cabling infrastructure. This met the requirements for spine-leaf network interconnection across equipment rooms. Compared with traditional single-mode DR modules, this solution significantly reduces cabling and deployment costs while ensuring signal transmission stability.

Application scenarios of Huawei StarryLink optical modules

FAQs

What Is a Separately Sold Optical Module?

In addition to being delivered with devices, Huawei's StarryLink optical modules can be purchased separately to meet the diverse deployment needs of AI and cloud data centers. Customers can select from a range of optical modules compatible with their existing network devices, accelerating resource expansion and upgrade cycles while improving O&M efficiency and inventory management.

The separately sold StarryLink optical modules strictly adhere to industry standards and support mainstream interfaces and protocols. They have also been validated through multi-vendor interoperability testing, demonstrating compatibility with major switch vendors and meeting AI cluster deployment requirements.

How Do I Select Optical Modules for Server Access and Switch Interconnection?

In typical data center scenarios, the proper optical modules to use between servers and access switches depend on the interface bandwidth of the server's NIC. For interconnection between access switches and aggregation switches, the proper optical modules are determined by the supported bandwidth of the switches and the actual service traffic model.

What Are the Key Specification Differences and Application Scenarios for VR, SR, DR, and FR Optical Modules?

The differences among VR, SR, DR, and FR optical modules are mainly reflected in the transmission distance, optical fiber type, and application scenario. The details are as follows:

Very Short Reach (VR): typically used for intra-device or short-distance connection.
Transmission distance: 50 m

Optical fiber type: multi-mode fiber (MMF)

Application scenario: interconnection between devices in a data center.
Short Reach (SR): commonly used in local area networks (LANs) and data centers.
Transmission distance: 100 m

Optical fiber type: MMF

Application scenario: short-distance high-speed transmission between devices in a data center and between switches.
Datacenter Reach (DR): A DR optical module is designed for data centers and high-speed communication scenarios. It features short-distance high-bandwidth transmission, and uses the parallel fiber technology to physically combine multiple optical signals into one optical fiber for transmission.
Transmission distance: 500 m

Optical fiber type: single-mode fiber (SMF)

Application scenarios:
1. Data Center Interconnect (DCI): It is widely used for high-speed interconnection between servers and switches, especially in large-scale parallel computing scenarios such as GPU clusters and AI training clusters.
2. Short-distance backbone network: It is used as a solution for short-distance high-bandwidth links on metropolitan area networks (MANs) or enterprise core networks, replacing traditional copper cables or long-distance optical modules.
3. Edge compute node: It provides compact high-bandwidth access for edge data centers, meeting requirements such as 5G base station backhaul and IoT data aggregation.
Forward Reach (FR): An FR optical module is designed for long-distance and highly stable point-to-point optical transmission. It usually works at a single wavelength and uses SMFs for data transmission. It uses the multiplexing technology to combine and separate signals using simple optical couplers or beam splitters, and combines optical signals of different wavelengths into different fibers of the same optical cable.
Transmission distance: 2 km

Optical fiber type: SMF

Application scenarios:
1. DCI: It implements efficient data exchange in high-bandwidth and low-latency scenarios across data centers.
2. Enterprise private line: It provides high-reliability and low-latency dedicated network connections for large enterprises.